Bruising pattern of table olives (‘Manzanilla’ and ‘Hojiblanca’ cultivars) caused by hand-held machine harvesting methods

Research Paper Bruising pattern of table olives (‘Manzanilla’ and ‘Hojiblanca’ cultivars) caused by hand-held machine harvesting methods Rafael R. Sola-Guirado a,* , Sergio Bayano-Tejero b , Fernando Aragon-Rodriguez b , Araceli Pe~ na c , Gregorio Blanco-Roldan b a Department of Mechanics, University of Cordoba, Campus de Rabanales, Cordoba, Spain b Department of Rural Engineering, University of Cordoba, Campus de Rabanales, Cordoba, Spain c Research Centre CIAIMBITAL, University of Almerı´a, Almerı´a, Spain article info Article history: Received 24 May 2021 Received in revised form 7 January 2022 Accepted 12 January 2022 Published online 28 January 2022 Keywords: Bruise index Colour Damage Fruit characterisation Mechanical harvesting Olea europaea This work presents a characterisation of the fruit and the bruising caused by some common detachment methods (manual, stick, shaker comb, branch shaker) and interception methods (net or padding) in common table olive varieties. We took pictures of fruit samples inside a special device, and the images were processed to extract characteristic parameters of shape and size (number of spots, Feret diameter, circularity, colours …). Moreover, we studied the time evolution of bruising caused on the fruit by a controlled impact. Finally, we developed a system that allows synchronised rotation of the fruit with image capture to evaluate bruising on the whole volume of the fruit. Our results showed that different harvesting treatments produced differences in the average number and diameter of spots per fruit, as well as in the average area of the spots per fruit for the different varieties. Fruit colour or bruising can also serve as a control factor for computer vision characterisation, for which reason we recorded differences in the firmness of the bruised and non-bruised areas of fruit. The harvesting method that caused the highest median values of bruise index was the shaker comb, particularly for ‘Manzanilla’ with an index of 1.59% on padding compared to 0.24% for ‘Hojiblanca’. Net interception was also observed to increase the bruise index in ‘Manzanilla’ (5.85%). Bruising assessment that only considers a single photograph means that a considerable amount of bruising remains disregarded compared to the actual bruising on the whole volume of the fruit. ©2022 The Author(s). Published by Elsevier Ltd on behalf of IAgrE. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). *Corresponding author. E-mail address: [email protected] (R.R. Sola-Guirado). Available online at www.sciencedirect.com ScienceDirect journal homepage: www.elsevier.com/locate/issn/15375110 biosystems engineering 215 (2022) 188e202 https://doi.org/10.1016/j.biosystemseng.2022.01.010 1537-5110/©2022 The Author(s). Published by Elsevier Ltd on behalf of IAgrE. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). 1. Introduction The cultivation of the olive tree (Olea europaea L.) enjoys sustained growth at a global level, currently reaching 10.5 Mha (FAO, 2019). Slightly less than half of the total production, 2.9 million tonnes, is destined for table olives. Spain plays a major role in the table olive sector, with a production of 0.5 million tonnes (IOC., 2020). The ‘Manzanilla’, ‘Gordal’ and ‘Hojiblanca’ cultivars (Campus, Degirmencioglu, &Comunian, 2018) are the most common commercial varieties due to their good pulp/pit ratio, and green processing is one of the most common methods employed for table olives. The Spanish style, which predominates in this method (S anchez-G omez, Garcı´aGarcı´a, &Garrido-Fern andez, 2013;Royal Decree 679/2016), means that any fruit defects require special attention as they influence quality and, therefore, the price fruit can attain on the market. Any spots on the fruit's surface will determine its category for commercialisation (Royal Decree 679/2016Royal Decree 679/2016). Such defects can be classified according to their origin (Riquelme, Barreiro, Ruiz-Altisent, &Valero, 2008), and bruising constitutes the characteristic type of damage produced during harvesting. Mechanical harvesting considerably increases the level of damage to fresh fruit such as table olives (Hussein, Fawole, & Opara, 2020). The main objective of the manual harvesting conducted in most table olive orchards is to reduce damage to the fruit. This causes a considerable loss of profitability since manual harvesting methods compete with the highly mechanised systems employed in modern olive growing for the new intensified planting systems. Advances in this field are oriented towards comprehensive mechanisation using combined harvesting machinery (Ferguson et al., 2010;SolaGuirado, Castillo-Ruiz, Blanco-Roldan, Gonzalez-Sanchez, & Castro-Garcı´a, 2020) and post-harvest systems (Rejano, Monta~ no, Casado, S anchez, &De Castro, 2010;Zipori, Fishman, Zelas, Subbotin, &Dag, 2021). In many farming contexts, mechanisation is not a valid solution due to the existing orography, the small size of the farm or a lack of resources, all of which make the introduction of machinery difficult. The only alternative on these farms is harvesting with semi-mechanised systems (Bernardi et al., 2018) such as branch shakers and shaker combs, or traditional methods like beating or hand picking, in which the fruit falls onto nets or suspended systems. These semi-mechanised harvesting systems detach the olive in a different way, so it seems logical to think they will also produce a different damage pattern. Similarly, since damage is related to the level of energy they receive (Jim enez-Jim enez, Castro-Garcı´a, Blanco-Rold an, Agu ¨era-Vega, &Gil-Ribes, 2012), there may be a relationship with the interception system, due to the correlation between damage and the height of the olive drop (Saracoglu, Ucer, & Ozarslan, 2011). Opara and Pathare (2014) list different procedures to characterise damage in different fresh fruit. Non-destructive techniques such spectral imaging, nuclear techniques, thermal imaging and ultrasound imaging, among others, have a great potential to determine the internal properties of fruit and detect bruising (Stella et al., 2015;Mohammed Raju, Jannat, Wang-Hee, Changyeun, &Byoung-Kwan Cho;Du et al., 2020). Computer vision techniques are more limited in that they can only characterise external damage to fresh fruit. However, for the characterisation of green olives that will be processed Spanish style (Campus et al., 2018), visible spectral imaging may be sufficient to perform bruise characterisation with traditional, low-cost technologies. Surface damage means any defect in the exocarp of a fruit, and may be associated with damage to the innermost layers, so damage might be visible and measurable (Li &Thomas, 2014). It is difficult to establish empirical formula for the quantification of bruising, which depends on different parameters such as variety and maturity status among others. A proper characterisation should take into account the amount, shape and colour of the damage. Bruise area is particularly suitable for the quantitative characterisation of damage. If the relationship between the bruise area and the overall fruit area is considered (Jim enezJim enez, Castro-Garcı´a, Blanco-Rold an, Gonz alez-S anchez, & Gil-Ribes, 2013) it is possible to establish a bruise index (BI). This classification allows for a quantitative qualification of the damage, without establishing discrete levels, as proposed by Hadi, Ahmad, &Akande, 2009. It is important to take special care to quantify fruit bruising over the total surface area of the fruit. (Corkidi, Balderas-Ruı´z, Taboada, Serrano-Carre on, & Galindo, 2006). The relationship of bruise index with the most commonly employed table olive harvesting systems would provide valuable information to improve the quality of mechanisation. Likewise, another parameter of interest could be the characterisation of the shape of damage in olives, in line with the considerations of other authors regarding bruise calculation in other fruits (Kitthawee, Pathaveerat, Srirungruang, & Slaughter, 2011;Mohammed et al., 2017;Van Zeebroeck et al., 2007). Image analysis can also determine the colour and ripeness of olives (Guzm an, Baeten, Pierna, &Garcı´a-Mesa, 2015). Other studies (Beyaz, Martı´nez Gila, G omez Ortega, &G amez Garcı´a, 2019) indicate the interest of applying similar methods to determine the colour of the type of fruit damage known as browning in the segmentation of olive damage. Bruising is changeable and requires the study of its evolution from its beginning (Jim enez-Jim enez, Castro-Garcı´a, Blanco-Rold an, Gonz alez-S anchez, &Gil-Ribes, 2013). To determine olive bruising quantitatively and qualitatively, visible imaging systems with computer techniques have shown great potential for delimiting the browned fruit areas after harvesting (Jim enez-Jim enez, Castro-Garcı´a, Blanco-Rold an, Gonz alez-S anchez, &Gil-Ribes, 2013;Beyaz, € Ozkaya, & _ Ic¸en, 2017;Ponce, Aquino, Millan, &Andujar, 2019). The aim of this article is to characterise the fruit and the bruise that occur in green table olives, specifically the ‘Hojiblanca’ and ‘Manzanilla’ varieties, employing computer vision techniques after the use of common harvesting methods. These parameters were evaluated when external damage had stabilised and, in addition, during temporal evolution. The comparison of damage quantification also took into account the entire surface of the fruit versus only one side of the fruit. biosystems engineering 215 (2022) 188e202 189 2. Materials and methods We conducted a series of trials, which consisted of harvesting olives according to different treatments and analysing the pattern of external damage caused to the fruits by means of image analysis. 2.1. Obtaining images to characterise the olives The olive samples were placed in 0.18 m square polylactic acid (PLA) trays, covered in blue foam with ellipsoid perforations for the placement of 20 fruits (Fig. 1). The samples were placed in a closed lighting device (Sola-Guirado, Bayano-Tejero, et al., 2020) with controlled lighting 13.95 ±0.57 lux, and a colour temperature of 5500 K from 4 LED bars on the floor at 30, with diffused lighting. A digital camera (Nikon, D80, Tokyo, Japan) was placed over of the samples, 570 mm above the tray surface, to take photographs at an aperture setting of f/6.3, shutter speed E:1/125 s, light sensitivity ISO:160, focal length FL: 35 mm, and exposure compensation EB:-1 EV. The resolution of each photograph was 3872 2592 pixels per tray area with a fruit resolution of 226 197 pixels. Each photograph was digitally processed by an expert using ImageJ software (National Institute of Mental Health) to characterise the fruit and its bruised area. To do this, the picture was first cut to obtain 20 individual photographs of each fruit, identified according to its treatment. Segmentation converted each photograph into a binary image to extract the fruit measurements. Then, by adding the binary photo to the original fruit photo, the background was removed, leaving only the fruit. Finally, the bruised area of each fruit was manually segmented to obtain another photograph with the bruising areas. Analysis of the pictures reported the following useful parameters for characterising the fruit and bruising pattern: - Fruit and bruise area (mm 2 ): the average number of pixels of each fruit or bruise scaled with its pixel-tomillimetre conversion. - Bruise index (%), BI: ratio between the total bruise area of each fruit and its fruit area. - Fruit length and diameter (mm). Largest and smallest distance between two points on the fruit area, usually coinciding with the major and minor diameters of the elliptical shape of the fruit. - Maximum Feret diameter (mm) of the bruise spot: longest distance between any two points along the spot selection boundary. - Circularity (#): the roundness or similarity to a perfect circle of the fruit or bruised spot, which varies from 0 to 1, with 0 corresponding to an infinitely elongated polygon and 1 to a perfect circle. - Number of spots: number of bruised spots per fruit. - Colour: coordinates on the average RGB colour space of the pixels of each fruit without taking into account the bruised area, or of the bruised spot itself. 2.2. Harvesting treatments of trees and olive samples We conducted the trials on two different plots of ‘Manzanilla’ table olives and another two plots of ‘Hojiblanca’ destined for green processing, located in Cordoba (Spain) during the months of September and October of the 2018/19 and 2019/20 harvesting campaigns. The trees were under irrigation and without biotic or abiotic stress. The harvesting systems used to detach the fruit were those habitually employed for this purpose: manual picking (M) as a reference treatment, manual beating with a long fiberglass pole (B), semi-mechanised with a branch shaker (BS) (Stihl, SP 481, Waiblingen, Germany), and semi-mechanised with a shaker comb (SC) (Pellenc, P230, Pertuis, France). The vibration signals produced by the harvesting methods on the main branches were recorded using a MEMS triaxial accelerometer (Gulf Coast Data Concepts LLC X200-4, Waveland, MS), with a measurement range of ±2000 ms 2 , a sensitivity of 0.06 m s 2 and a sampling frequency of 400 Hz. Table 1 summarises the vibration patterns that characterised these systems on the branches. Each tree was harvested from branches located between 1.5 and 2.4 m above the Fig. 1 eTray with impacted fruit and sequence of impact caused by free-fall of the ball. biosystems engineering 215 (2022) 188e202190 ground by a farm worker, using the usual technique employed on the plot. An agricultural textile netting (N), as usually used in olive harvesting, was placed on the floor, along with a padding surface (P), consisting of two layers of 3 mm thick polyvinyl chloride filled with air with a diameter of 1 m at a height of 0.1 m. Olive samples were taken from the padding to study the exclusive incidence of the detachment method. In the case of the ‘Manzanilla’, variety, which has greater susceptibility to bruising (Jim enez-Jim enez, Castro-Garcı´a, Blanco-Rold an, Gonz alez-S anchez, &Gil-Ribes, 2013), samples were also taken from the net in areas far from the padding to study the incidence of the interception means on this variety. In addition, randomly hand-collected samples without external damage were harvested and an impact or hit (H) was applied to the centre of the fruit. To perform the impact, a device was used in free fall from a height of 0.125 m with a steel ball that had a mass of 0.035 kg, applying an energy of 0.043 J (Fig. 1). The following treatments were thereby obtained: MP, BS-P, SC-P, BeP for ‘Hojiblanca’ and MN, MP, BS-N, BS-P, SC-N, SC-P, BeN, BeP for ‘Manzanilla’ and H for both varieties. The samples obtained in each of the treatments, were kept dry at an average temperature of 23 C and photographed 150 min after harvesting, when the bruise had stabilised (Jim enez-Jim enez, Castro-Garcı´a, Blanco-Rold an, Gonz alez-S anchez, &Gil-Ribes, 2013)., avoiding those that circumscribe the perimeter of the plot. Three rows of trees were randomly selected and, in each row, three trees were again randomly selected to apply the treatments (1 tree exclusively used for 1 harvesting method). When selecting rows, those corresponding to the perimeter of the plot were discarded, as were the trees at the beginning and end of the row. Twenty fruit samples (1 tray) were taken from each tree for each treatment (BSeP, SC-P, BeP, BS-N, SC-N, BeN), resulting in a total 60 samples per treatment and day. Treatments with a common harvesting method were applied on the same tree, and samples were harvested simultaneously using the net or padding surface. For the reference treatment (M), olives were randomly harvested from the 3 trees in the row before application of the treatments. In case of impact treatment (H), I olive samples were also collected from the 3 trees for the treatments in the row, on different areas of the tree and without previous bruising. These trials were repeated 4 times per campaign and plot, spaced approximately 5 days apart, leaving a minimum difference of 4 days in the event of rainfall. 2.3. Characterisation of olives and their external damage We conducted different types of studies to determine useful parameters for modelling olives and their bruising: 1. Size and shape characterisation: Several geometrical parameters were taken for the fruit (length, diameter, and circularity), for the bruised spots they had (number of spots, Feret diameter and circularity) and for all treatments, extracting the information from image analysis. In addition, the fruit mass (g) was measured with a digital scale (Gram, EH-500, Spain). 2. Firmness assessment: The penetration force needed to break the fruit surface was measured with a penetrometer using a cylindrical 3 mm long and 2.4 mm diameter tip (IMADA Inc., DS2-11, USA) in non-bruised and bruised areas (the latter from spot areas greater than 16 mm 2 ). 3. Colour determination: The colorimetric characteristics of the olive in fruit areas with damage (spot area greater than 16 mm 2 ) and without damage were measured for all treatments using a colorimeter (Konica Minolta, CR400, USA) calibrated with a D65 2illumination measured in CIELAB colour space. Similarly, we determined the RGB colour coordinates of the images processed. 4. Bruise index and its time evolution: The bruise index was determined 150 min after harvesting for all treatments, using image analysis to extract the fruit area and damage area. Moreover, images of the fruit impacted with the free-fall device (H) were taken every 15 min up to 150 min, to evaluate the development of bruising over time. 5. Location of the bruising on the surface of the fruit: Calculation of the bruise index normally uses a single image, ignoring what occurs in the unseen areas, considering the ellipsoid geometry of the olives. To evaluate fruit bruising considering the entire surface of the fruit, we designed a prototype (Fig. 2) to rotate the olives on their main axis. The device consists of several gears, with two needles in their centres, where randomly selected fruit samples are placed. We evaluated 60 fruit samples per harvesting treatment and variety, 150 min after their harvest. A motor (4076 steps per turn) controlled by a microcontroller (Arduino, Nano, Italy) rotated the fruit, shifting 15at a time to take 24 pictures per fruit in each position of the revolution. To determine the amount and location of the bruising produced over the whole surface, the pictures were processed to give a complete two-dimensional representation of the olive without deformation (Fig. 3): a. Each of the captured images was cropped to avoid overlapping of the same area, circumscribing an ellipse whose major diameter coincided with the fruit length Table 1 eVibration parameters of fruit-bearing branches with harvesting systems. Harvesting method Mean resultant acceleration (m$s 2 ) Frequency (Hz) Stick beating 560.7 ±214.3 a e Shaker Comb 79.6 ±42.6 14.0 ±0.4 Branch shaker 209.3 ±81.1 20.6 ±1.1 Values showed are mean ±standard deviation. a Mean peak value. biosystems engineering 215 (2022) 188e202 191 and whose minor diameter is the fruit perimeter ðp∙diameterÞdivided by 24. b. Each cut was joined to the consecutive one. c. All pixels of the composition are grouped together to give continuity to the composition while maintaining the equatorial line constant. d. The remaining composition was enlarged by projection onto the curvature of the ellipse in the front view, i.e. by making the distance between the peduncle and its antipode coincide with half the perimeter of the ellipse determined in the front view. In all cases, we investigated the relationship that exists for all of the studied variables with both the harvesting method and the interception method (in the case of ‘Manzanilla’). We also analysed the differences between the different varieties before finally studying relationships existing within the different study parameters. 3. Results 3.1. Fruit and fruit bruising size and shape characterisation Table 2 shows the characterisation of shape and size of the fruits sampled for the two harvesting seasons according to the different methods. Fruit mass, length and diameter variables showed a normal distribution (KolmogoroveSmirnov, p >0.05) for both varieties. A comparison of the harvesting methods (ANOVA test, p >0.05; post-hoc Tukey test, p >0.05) showed no significant differences in the variables of mass, length and diameter for the two varieties and seasons, nor did we find any significant differences between the net and padding catching systems (ANOVA test, p >0.05; post-hoc Tukey test, p >0.05). However, we did find significant differences (Student'st-test, p<0.05) within each variety in the comparison of the 2019 and 2020 seasons: The diameter and length of fruit for both varieties showed a strong positive correlation with fruit mass (‘Hojiblanca’: Pearson Coefficient ¼0.88; Pearson Coefficient ¼0.89; p<0.05; ‘Manzanilla’: Pearson Coefficient ¼0.88; Pearson Coefficient ¼0.91; p <0.05). Circularity did not show a normal distribution (KolmogoroveSmirnov, p <0.05). Significant differences were found in circularity between varieties (WilcoxoneManneWhitney Test, p <0.05), with a median value and interquartile range in ‘Hojiblanca’ of 0.865 (0.023) and 0.869 (0.026) in ‘Manzanilla’,with diffuse differences between harvesting methods. Table 3 shows some representative parameters of the characteristic size and shape of the fruit bruising spots. The number of spots per fruit did not follow a normal distribution. For both seasons and varieties, the number of spots was significantly different (KruskaleWallis, p <0.05) between the manual method and the other detachment methods (post-hoc Wilcoxon rank sum test with Holm adjustment, p <0.05), in which there were no differences between them (post-hoc Fig. 2 eDevice developed for turning olives and taking images from different sides. biosystems engineering 215 (2022) 188e202192 Wilcoxon rank sum test with Holm adjustment, p >0.05). In all cases, the number of spots within the same detachment methods increased in the net catching compared to padding catching. The circularity of spots was significantly different between detachment methods for ‘Manzanilla’ and ‘Hojiblanca’ (KruskaleWallis, p <0.05; post-hoc Wilcoxon rank sum test with Holm adjustment, p <0.05), and was lower for the latter in all cases. The mean greater Feret diameter of the spot was different (KruskaleWallis, p <0.05; post-hoc Wilcoxon rank sum test with Holm adjustment, p <0.05) between varieties and detachment methods, and slightly greater in ‘Hojiblanca’, except for the shaker comb, where they were the same. 3.2. Firmness assessment Penetration force values in the bruised and non-bruised fruit zones followed a normal distribution (KolmogoroveSmirnov, p>0.05) for both varieties. Significant differences (Student'sttest, p <0.05) were found between the bruised and nonbruised fruit zones in both varieties, with a value of 14.02 ±1.92 N and 10.26 ±1.73 N for the bruised zone in ‘Hojiblanca’ and ‘Manzanilla’, respectively, and a value of 17.11 ±1.73 N and 13.02 ±1.32 N for the non-bruised zones, in '‘Hojiblanca’ and ‘Manzanilla’, respectively. Within each variety there were also significant differences between the Fig. 3 eProcess performed for the representation of the entire surface of an olive from the 24 photos of a fruit rotated 15. Table 2 eFruit size and shape values measured in two harvesting seasons according to detachment method with padded catching and olive variety. Detachment method Variety Mass (g) a Length a Diameter a Circularity b Manual ‘Hojiblanca’ 3.64 ±0.84 a 22.42 ±2.01 a 17.71 ±1.49 a 0.869 (0.024) a ‘Manzanilla’ 4.22 ±1.00 b 22.72 ±2.27 abc 18.62 ±1.86 b 0.877 (0.024) b Branch shaker ‘Hojiblanca’ 3.58 ±0.83 a 22.43 ±2.02 a 17.60 ±1.56 a 0.864 (0.021) c ‘Manzanilla’ 3.91 ±0.98 c 22.92 ±2.11 bd 18.47 ±1.63 b 0.866 (0.026) ac Shaker comb ‘Hojiblanca’ 3.56 ±0.70 a 22.36 ±1.64 a 17.68 ±1.35 a 0.863 (0.024) c ‘Manzanilla’ 3.90 ±0.97 c 22.59 ±2.13 ab 18.25 ±1.78 b 0.867 (0.025) a Stick beating ‘Hojiblanca’ 4.78 ±0.86 d 23.32 ±1.83 d 19.38 ±1.41 c 0.878 (0.021) b ‘Manzanilla’ 4.74 ±0.79 d 23.16 ±1.62 cd 19.35 ±1.34 c 0.879 (0.022) b Values shown are mean ±standard deviation or the median and the interquartile range in brackets. Different letters between rows of the same column indicate significant differences according to. a Normal distribution (ANOVA, p <0.05; post hoc pairwise t with pooled standard deviation and Holm adjustment method, p <0.05) or. b Non-normal distribution (KruskaleWallis, p <0.05; post hoc Wilcoxon rank sum test with Holm adjustment method, p <0.05). biosystems engineering 215 (2022) 188e202 193 bruised and non-bruised zones (paired t-test, p <0.05), with an average reduction in penetration force of 18% for ‘Hojiblanca’ and 21% for '‘Manzanilla’. In terms of harvesting methods, the fruits of both varieties followed the same trend in the non-bruised zones. There were significant differences of penetration force between the manual method, which had the highest value, and the other harvesting methods (ANOVA test, p <0.05; post-hoc Tukey test, p <0.05). In the non-bruised zone, no differences were found between the branch shaker and shaker comb for ‘Hojiblanca’ (Student's t-test, p >0.05), with values of 14.91 ±1.68 N and 14.10 ±1.94 N, respectively, whereas differences were found between these methods for '‘Manzanilla’ (Student's t-test, p <0.05), with values of 11.82 ±1.05 N and 9.73 ±1.61 N. A positive correlation of fruit penetration force with fruit weight (Pearson coefficient p <0.05) was found for both varieties in manual harvesting methods, considering that the fruit measured are destined for green processing. Penetration force values in the non-impacted zone of the fruit showed variability over time. Significant differences were found in both varieties between the first four weeks and the fifth and sixth weeks (ANOVA, post-hoc Tukey p <0.05; Kruskall-Wallis, post-hoc WilcoxoneManneWhitney Test p<0.05). We observed that the proportion of fruits with a higher maturity (higher Jaen index), increases as weeks go by (data not shown). In addition, the value of their penetration force was different among fruits, except for those with indices of 0 and 1 (ANOVA, post-hoc Tukey p <0.05; Kruskall-Wallis, post-hoc WilcoxoneManneWhitney Test p <0.05), which decreased in value as the maturity index increased. 3.3. Fruit and fruit bruising colour characterisation Table 4 shows the components of CIELAB colour space measured with the colorimeter on the fruit samples tested. The values of the ‘a’ component showed no differences between the different campaigns for each variety (WilcoxoneManneWhitney Test, p >0.05), unlike the ‘L’ and ‘b’ components, which were different for each variety (WilcoxoneManneWhitney Test, p <0.05). In addition, we observed no significant differences between the different mechanised harvesting methods or between the padding or net interception methods. The values of all components were higher in ‘Manzanilla’ than in ‘Hojiblanca’ in both seasons. In both varieties, negative relationships were measured between the ‘a’ and ‘b’ component of the CIELAB colour space (Spearman rho, p <0.05) with significant relationships related to fruit mass and diameters (Pearson coefficient, p <0.05; Spearman rho, p <0.05) for ‘Manzanilla’. Although this study only evaluated green fruits, when considering different maturity indices, the analysis of the fruit colour in RGB space extracted from the photographs shows a trend of colour evolution from deep green to black. (Fig. 4). Fruit colour over the six different weeks studied was significantly different (Kruskall-Wallis, p<0.05) for each of the channels (RGB), with different red and green channel values for each week (post-hoc WilcoxoneManneWhitney Test with Holm correction, p <0.05). 3.4. Bruise index characterisation Bruise index 150 min after harvesting The bruise index did not follow a normal distribution (KolmogoroveSmirnov, p <0.05). Figure 5 shows the median values obtained for the two varieties and the harvesting treatment studied. There were significant differences (WilcoxoneManneWhitney Test, p <0.05) for the two seasons between ‘Hojiblanca’ and ‘Manzanilla’ with greater damage suffered by ‘Manzanilla’. Considering exclusively detachment method, i.e. comparing padding treatments, there were no significant differences between varieties using the manual method and manual beating with stick (KruskaleWallis, p<0.05; post-hoc Wilcoxon rank sum test with Holm adjustment, p <0.05). However, significant differences were found between varieties using the branch shaker (BI ¼0.03 for ‘Hojiblanca’ and BI ¼0.33 for ‘Manzanilla’) or the shaker comb (BI ¼0.24 for ‘Hojiblanca’ and BI ¼1.59 for ‘Manzanilla’), with significant differences between both methods for the same variety (post-hoc Wilcoxon rank sum test with Holm adjustment, p <0.05). Regarding the intercept method, i.e., comparing the same detachment method between padding and net, we found differences for all systems, which were Table 3 eFruit bruised spot size and shape values measured in two harvesting seasons according to harvesting method and olive variety. Detachment method Catching method Variety Number of spots Feret Diameter (mm) Circularity Manual Padded ‘Hojiblanca’ 0 (1) a 2.35 (2.10) a 0.709 (0.235) ab Padded ‘Manzanilla’ 0 (1) a 1.62 (2.20) b 0.788 (0.314) c Net 2 (3) b 2.21 (2.01) ab 0.674 (0.233) ab Branch shaker Padded ‘Hojiblanca’ 1 (2) c 1.53 (1.10) b 0.725 (0.205) a Padded ‘Manzanilla’ 2 (5) b 1.34 (0.83) c 0.787 (0.191) c Net 3 (4) d 1.99 (1.42) b 0.704 (0.198) ab Shaker comb Padded ‘Hojiblanca’ 1 (2) c 3.04 (2.60) d 0.628 (0.211) d Padded ‘Manzanilla’ 2 (3) b 2.79 (2.16) ad 0.677 (0.201) b Net 3 (2) d 4.41 (2.37) e 0.580 (0.162) e Stick beating Padded ‘Hojiblanca’ 0 (2) ac 2.08 (2.61) ab 0.666 (0.379) abd Padded ‘Manzanilla’ 0 (3) ac 2.37 (2.18) abd 0.664 (0.328) abd Net 2 (3) b 2.28 (2.34) abd 0.628 (0.305) abd The values represented are median and interquartile range. Different letters indicate significant differences between rows of the same column (KruskaleWallis, p <0.05; post-hoc Wilcoxon rank sum test with Holm adjustment method, p <0.05). biosystems engineering 215 (2022) 188e202194 significantly higher with a net intercept (KruskaleWallis, p<0.05; post-hoc Wilcoxon rank sum test with Holm adjustment, p <0.05). The highest values were found using the shaker comb for both seasons and varieties with greater values in ‘Manzanilla’ (BI ¼1.59 for padding and BI ¼5.85 for net). It is possible to make a nominal classification by placing bruise index limits (BI ¼0, 0<BI <1, 1<BI <3, 3<BI <5, BI >5) in order to compare the level of damage studied by other authors (Castro-Garcia, Castillo-Ruiz, JimenezeJimenez, GilRibes, &Blanco-Roldan, 2015: no damage, slight damage, moderate damage, severe damage, mutilated). However, this is only a proposal of thresholds since these authors used a visual classification so did not quantify the area of bruising used to define the limits. Figure 6 shows the distribution of bruising in the sampled fruit with our proposed classification. With the thresholds, we obtained a distribution of the level of damage caused by each harvesting method studied (Fig. 7). Bruise index and its development over time Figure 8 shows the temporal evolution of the bruise index from the time the controlled method (H) impacted the olive. Bruise index values at different times post impact showed significant differences (Student's t-test, p <0.05; WilcoxoneManneWhitney Test, p <0.05) between varieties. The median bruise index values were fitted to a logarithmic Table 4 eCIELAB colour space measured with the colorimeter on fruit samples impacted with the controlled energy method (H) and those harvested with the shaker comb, branch shaker and beating methods using padding in the interception. Detachment method Variety L A b Controlled energy (H) (damage zone) ‘Hojiblanca’ 52.9 ±2.9 c 15.1 ±1.7 b 32.4 ±3.2 c ‘Manzanilla’ 52.6 ±3.8 c 11.5 ±2.9 c 30.3 ±3.8 d Controlled energy (H) (undamaged zone) ‘Hojiblanca’ 58.6 ±3.0 a 19.2 ±1.1 a 38.6 ±2.3 a ‘Manzanilla’ 62.1 ±2.5 b 18.8 ±3.2 a 40.8 ±3.5 b Manual ‘Hojiblanca’ 59.5 ±3.0 a 18.6 ±2.6 ad 37.8 ±2.8 a ‘Manzanilla’ 62.4 ±2.3 b 18.5 ±2.0 d 39.4 ±3.0 e Mechanical harvesting (SC, BS, B) ‘Hojiblanca’ 59.0 ±3.1 a 19.0 ±1.3 a 36.5 ±4.6 f ‘Manzanilla’ 61.9 ±4.5 b 18.3 ±4.0 d 38.4 ±2.9 a Values shown are mean ±standard deviation. Different letter indicates significant differences between rows of the same column (KruskaleWallis, p <0.05; post-hoc Wilcoxon rank sum test with Holm adjustment method, p <0.05). *The colorimeter was placed on a random area of the fruit regardless of whether the area was damaged or not, except for the controlled energy method, in which it was positioned just above the area impacted by the ball. Fig. 4 eColour evolution (R) during table olive ripeness throughout the harvesting season (median of the values in RGB space for Hojiblanca y Manzanilla). (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) Fig. 5 eBruise index for each treatment of fruit detachment (M: manual by hand, BS: branch shaker, SC: shaker comb, B: manual beating) and interception (N: nets, P: padding) and for each olive variety. The values shown represent median and interquartile range (KruskaleWallis, p <0.05; post-hoc Wilcoxon rank sum test with Holm adjustment method, p <0.05). biosystems engineering 215 (2022) 188e202 195 function in relation to the time post impact. Figure 9 illustrates the colour differences of the bruised spots between the RGB colour channels over time. Colour differences between the two varieties of fruit over the time were found in the red and green channel (Friedman Test p <0.05) but not in the blue channel. The greatest changes over time occurred in the green channel, where values decreased with a greater slope in ‘Manzanilla’. Location of the bruising on the surface of the fruit Figure 10 shows the differences in the bruise index between analysing a single image that partially captures the surface of the fruit, and the proposed system, which captures the entire fruit surface. The methods based on branch shakers and manual harvesting showed significant differences between the bruise index determined from a single photo or from the whole perspective in true magnitude. Conversely, harvesting methods with shaker comb and beating showed no significant differences. In general, when the image of the spread external surface of the fruit is considered, there is a marked increase in the percentage of bruising, a median 140%, considering that 100% is the estimated bruising from a single unprocessed zenithal picture, although a high deviation exists in some cases. Fig. 6 eDistribution of bruising values, boundary lines of each damage category shown with vertical dotted lines. Fig. 7 eFrequency of damage category for each treatment of fruit detachment (M: manual by hand, BS: branch shaker, SC: shaker comb, B: manual beating) and interception (N: nets, P: padding) and for each olive variety. Different letter indicates significant differences between treatments in the same damage category (pairwise two-samples Z-test of proportions with Holm adjustment method, p <0.05). biosystems engineering 215 (2022) 188e202196